After doing a major wholesale MAME update once before and finding it to be a timesink of dubious value, I decided that I needed a different strategy going forward. So here's my 2 cents on the upgrade thing. There have definitely been worthwhile improvements since 146, especially in sound, as well as some new games. As everyone knows though, MAME upgrades not only bring improvements, but often seemingly random slow-downs and breakage. So I never find myself doing wholesale updates anymore. Rather, all my MAME upgrades are done alongside my "base" install (which happens to be 146). In other words, new mame revs are just installed in parallel folders (such as MAME153). When I discover that a particular game has been recently improved or a new one has been added, I just copy the rom into the rom folder of the parallel MAME folder, and that's it. Instant upgrade for just that one game. If I don't like the way the new MAME ran the game (and that does happen), I just delete the rom and its back to running the old way (The way this magic works is that the FE doesn't launch MAME directly, but runs a simple AHK script which searches for the rom in all MAME* folders from newest to oldest and executes it using the first MAME rev it finds it in).
When yet another rev of MAME comes out, I usually start by copying all the roms (and control files) from the previous rev to the latest (say, from MAME 153 to MAME168) and test each one. Since there is only a handful of roms to test (about a dozen--the rest still run under 146), it doesn't take long to do this. Usually if something doesn't work, its because I need a newer rom. If they all work, then I can delete the entire previous rev (say MAME153). Only if this collection of roms went over a hundred or MAME introduces some major new architecture would I think of doing a wholesale upgrade. I expect that to take at least a decade or maybe never.
I find this method works exceptionally well and avoids almost all the downsides of wholesale MAME upgrades (getting new rom sets, discovering arbitrary breakage the first time you run a game six months later, difficulties in rollback, etc.). Given the relatively slow rate of improvements to the classic game sets these days, I find this method to provide the most tactical way to take advantage of the latest MAME improvements without the headache.