In AD&D, they were generally just left there. In 3e, the game got one big revision. In 4e, it got a constant trickle of errata and the occasional 'feat tax.'
None of these are ideal. Simply leaving the problems is obviously sub-optimal. And I for one found the constant patching of 4e to be a major distraction, and detracted a great deal from my enjoyment of the game for little benefit.
IMO, the 3.0 -> 3.5 model was the best of the three... but that doesn't mean I consider it a
good thing, merely the best of a bad bunch. 3.5e certainly introduced its fair share of problems; many of the so-called 'improvements' were nothing of the sort.
(Oh, incidentally, 3e got at least
two big revisions - the revision to the polymorph rules was pretty huge.)
Should 5e publish frequent errata as problems are identified? Kludge problems with feat taxes and replacements that obviate existing inferior elements? Have a '5.5' to fix problems identified in 5.0? Or just leave it to the DMs to tweak anything that's a problem 'for them?'
A combination.
For the most part, they should release new material initially through the DDI. Then, over a period of time they should be able to identify the issues and apply patches.
Once they're reasonably sure that they've fixed the issues, they should compile the new material into the print version,
and then leave it alone. IMO, they need to have the discipline to refrain from revising material once it's in print. (True errata is a slightly different beast - that usually consists of correcting typos. But many of the 4e changes went way beyond simple errata, and were true revisions; it is those that were most objectionable.)
Where they find that some new option means that an older one is suddenly obselete or too powerful, they need to adopt the policy of fixing the
new material. Again, once in print the older stuff should be considered 'locked'.
However... no matter how careful you are with getting material into print, and no matter how disciplined you are with revisions, you will eventually come to a point where something simply
cannot go unchanged any longer. Perhaps some power has been written so that
any expansion in that area becomes impossible without revising it, or perhaps some feat just needs removed from the game.
But that's fair enough, and is actually something they should embrace. So, a suitable time after the edition is first released, they should do "5e Revised" (or "5.5e", or 6e, or whatever), collecting the accumulated wisdom of the previous few years, applying those revisions they've wanted to make but been unable to do so (because the material was "locked"), and perhaps adding or removing some elements to the core. In effect, this would be the same notion as the 3.5e books, although I wouldn't advocate making changes of that scope again short of a full new edition.