Echoing Gerhard Paseman's comment more-or-less: it certainly never hurts to ask permission, or at least to explicitly express interest in use of on-line resources. Some on-line resources have indications of restrictions or lack thereof (e.g., Creative Commons License). In all cases, I think one should acknowledge on-line resources just as one would acknowledge traditional paper-media resources. Credit where credit is due.
Another point, more practical, is that the on-line versions may get updated, corrected, and revised, while downloaded (or printed) versions will certainly not. Thus, there is the "just in time" ("JIT") mechanism in play for on-line resources, namely, not to download (or print) until the last possible moment.
... and echoing anon, as well.
Edit: The concern about depending upon on-line work that later proves flawed should also be directed at "refereed journal" publications, because many journals nowadays explicitly tell referees that they are not responsible for determining correctness! It is supposedly the author's responsibility! Thus, if there's no scandalous outcome, perhaps no one has seriously attempted to corroborate.
And, yes, one can argue that things are put on-line to be used. But, still, I think credit should be given, just as with physical books and papers. E.g., just because I've purchased a book doesn't mean I don't owe an intellectual debt to the author, and to the people cited in the book. I think it is completely comparable to say that one should not, by failing to say anything to the contrary, give the impression that ideas one gleaned from someone else's writing are one's own. Being on-line or not does not suspend the rules of responsible, ethical scholarly conduct.
Other discussions in this and other forums have attempted to distinguish "logical dependence" from "acknowledgement of prior art". I have known some otherwise-reputable mathematicians who have explicitly said that since they deliberately avoid looking at other peoples' officially published work, they never have to mention it, perhaps despite having heard informal discussions of it. Or, if there's no logical dependence on someone else's stuff, it doesn't have to be mentioned, supposedly. I realize others will disagree, but I think "prior art" deserves as much mention as anything else. All the other things can be dodged, if one tries hard enough! I think we should respect prior art, as well as noting technical-logical dependence, and I do not think that on-line or not makes much difference in this.