Is it usual for a referee to heed updated versions on arxiv? I've put a paper on arxiv one year ago and I've submitted the version 6 to a journal seven months ago.  
During these last seven months, I've given several talks about this work, which led me to update again the paper on arxiv, to fix some typos, polish, add clarification, add a few new applications and questions... nothing very serious, but the last arxiv version (of 51 pages) is surely better than the submitted one (of 47 pages). 
Out of respect for the work of the referee and to not delay the report, I believe (perhaps wrongly) that it is better to not update the submission before the first report; but I was wondering:
Question:  Is it usual for a referee to heed updated versions on arxiv?
 A: As a referee, I usually stick to the submitted version, which is what I am asked to evaluate.
Often, it happens to encounter an error, an imprecision, or an unclear passage. Sometimes, it is not immediately clear whether it is just a typo, or something the authors really did not consider. In these cases, I usually check for a more recent version on the arXiv, or on the author's webpage, to see whether that part has been somehow improved or corrected.
So, updating papers on the arXiv even after submission could be beneficial to both authors and referees, by saving some time. 
Indeed the possibility to improve the manuscript after the submission is not an excuse to send out poorly revised manuscripts for evaluation.
To comment Friedrich Knop's answer, in my opinion it is a very good (but not widespread) practice to update the arXiv version to fix or improve parts of the manuscript - even after publication. Most of the times, however, the arXiv version is not even updated to reflect the (sometimes substantial) modifications suggested by the referees.
A: I think Friedrich Knop's answer describes a beautiful ideal, but it is not very realistic. For example, the following have both happened to me personally:


*

*I posted a preliminary version of a paper to the arXiv; later, after completing the paper and improving the results, I submitted the completed paper to a journal without updating the arXiv version. When I received the referee report, it was clear that the referee downloaded the version from the arXiv rather than reading the submitted version (they complain about the presence of hypotheses that don't appear in the paper I submitted, and refer to propositions with the numbering from the arXiv version). Milder versions of this story have happened with other papers.

*For another paper, I posted to the arXiv and submitted to a journal at the same time. Many months went by without a referee report. At some point I updated the version on the arXiv to reflect some minor corrections. When I received the report, the referee points out that some of their suggestions for changes to the paper have already been implemented in the more-recent arXiv revision.
In light of the first story, whenever I submit a paper, I now always make a point to update the arXiv version to be at least as current as the submitted version. In theory, perhaps this should not be necessary; but referees are human, and like everyone they sometimes take shortcuts.
A: It is not only unusual, I would be surprised if that ever happens. A referee's task is to a evaluate a paper in the state it has been submitted. So he/she should not take into account any other versions found elsewhere even if that means that the paper is rejected for poor quality. For that reason it is expected to only submit papers which are considered to be finished. In the case a serious problem is found after submission (like a gap in a proof, not just bad exposition) then the proper procedure is to inform the editor and resubmit the paper with profuse apologies.
A: I am hesitant to try to add anything further to the solid answers here, but here it goes anyway.
In practice, here is what I do:


*

*Once I have a manuscript that I feel is publisher-ready, I send it out to experts that I feel might be interested.  After a week or two I re-read, make updates, and incorporate any feedback I received.

*Then I post the manuscript on the arXiv.  I again wait a few weeks to see if any feedback comes.  After that period I make changes as appropriate and give it another read making edits as they arise.

*Then I submit to a journal.

*Generally speaking I do not update the arXiv between submission and a referee report.  In that time I always find typos and minor edits.  Sometimes new content wants in too.  To address this I make said changes and keep a log, commented out in TeX, at the top of my file, itemizing all changes that I have made.  After the referee report, I address the referee suggestions and if rejected update the arXiv and resubmit.  If accepted, I use the log I created to walk the referee through the changes (to make his/her job easier), and then update the arXiv.


This process has worked for me so far. There have been an occasion or two where I have broke protocol (exceptions always arise), but this is my attempt at a "best practice" from the author's point of view.
As a referee, I only consider the submitted manuscript.  If I were to receive an update in the middle, generally speaking, I would consider the "clock" reset and start over from scratch.  That said, I will sometimes compare the arXiv version to the submitted version.  But I evaluate exclusively on the submitted version.
