My comments above formulated as an answer:

People typically post a preprint on the arxiv at the same time that they post it on their own homepage, with the goal of disseminating their work to their colleagues. (These days, posting on the web is more important than journal publication for sharing your work, and the arxiv is the central repository for math on the web.) This is often at the same time that they submit to a journal, although sometimes they wait for feedback (as Joe Silverman suggests in comments above), and sometimes they spend more time polishing their preprint before submitting it (as Darij Grinberg suggests, again in the comments above).

The conclusion seems to be that it is standard to post on the arxiv as soon as one is ready to share one's work with colleagues, and that this is often at, or close to, the same time that one is ready to submit one's paper. In particular, it is quite common to post on the arxiv at the same time as submitting, or not long prior to submitting. (But there is nothing wrong with posting on the arxiv and then spending some more time polishing your preprint before submitting it to a journal.)