Have you looked at the other packages out there, and at Subversion? You might be pleasantly surprised to discover that many others have shared your pain and attempted (as you observed) that Bugzilla's complexity made it difficult to maintain without starting from scratch.

Sometimes you have to expect to throw away the prototype :-). Oh, wait, ALL THE TIME you have to expect that (cf. Fred Brooks).

Take a look at the Trac project sometime, it would benefit from your critical eye and your experience. I came from Mantis and promptly began hacking on my Trac installation (right into the core mod_python handler -- I admit it, I like to fix things NOW).

Ideas like AJAX integration (below) and better SVN/DB integration (above, though Trac already is tightly integrated with SVN) are more easily implemented in a decoupled framework. The critical difference I see between Bugzilla and Trac et al. is the decoupling of the latter, more modern systems. Having proper templating (although I find ClearSilver somewhat annoying, it's not hard to eg. add prototype.js to the headers, or use AJFORMS for form processing) is a huge plus.

Consider whether Subversion's changeset-centric versioning model might benefit your organization. It should not be overly difficult to migrate your existing hooks to SVN's hook system (almost trivial in fact), and the Perforce-like changeset notion makes groups of errors easier to track down (per-changeset). "Gee, this started happening around 11/19/05... what changed that day? And who submitted it? (narrows down time window)" ... "Oh for fuck's sake, THAT broke the config?!?"

Anyways. Bugzilla has too much baggage. It was the prototype. I submit to you, that the current generation has built upon its strengths and tried to address its weaknesses, and those who are just starting a project yet want a good OS solution, should consider Trac. You might want to give it a look too, if only to see whether there are any ideas they added that work for your organization's Bugzilla creation.