So hopefully, that limiter would only be acting on that 1 rogue transient.
Not if it's across the 2-buss... Find out what's causing it and put the limiter there.
Strapping a limiter across the mix buss may only by it's settings affect a rogue transient, but it's still sending the entire mix through the limiter, lengthing the signal chain and and moving the mix just that much further from the original truth.
Granted, in digital it may not make as much of a difference as in analog, but it's still there. Not to mention that using an entire limiter just to stunt a single transient or two is rather like using ariel napalm to remove a few weeds from the backyard; it's kind of overkill (even if it does smell good in the morning.)
I'm with John in that it's better to attack those in the tracks than on the mix buss. Where I'd personally ddiffer with all respect, though, is that I still would not use a limiter just for a couple of rouge waves, even on the source tracks. There are a few reasons I say this. First is the napalm on dandilions effect is still there. But more important, is that a limiter may be inappropriate in a case likeyou say where you say you can't find the cause.
Often times, one or two rogue transients will pop up on the mix buss when there is no apparent source cause for them on any one individual track. What is often the case is that there are two or three (or maybe more) perfectly innocous "peaks" on two or three (or more) individual tracks whose attacks just so happen to line up so perfectly on the timeline, have complimentary harmonic content, etc. so that when they are summed, they create a larger-than-average transient on the mix buss. The individual peaks may not even rise above the average peak level on the individual tracks; it's just the coincidence of their summingthat causes the rogue peaks.
In such cases, trying to limit the individual tracks would not work well, because you'd have to limit more than just one peak, you'd be limiting the average peak level for the entire track, which may not be what you want sonically. Instead I'd just go in on two of the most obvious of the source tracks and manually knock down that one mini-peak on each track by a couple of dB. The end effect will be inaudible on the track level, and on the mix buss, your rogue peak will be tamed by several dB just as if it were limited, but without the extra baggage.
G.
If I still like the mix, but just want to sweeten it as a whole, then I work on copies of the mix file, preserving the original mix file. I may try a few different approaches, outputting a premaster file for each, to review later, saving each session. It may be only when I assemble the CD as a whole that I choose which of those goes into the burn file. But, you know, that's just me. the right methodology is the one that makes you do your best work.
. For example, to call the pancaking of that Radiohead album "amazingly transparent" sounding is kind of like saying that they were suprised when they drank a glass of molton lava of just how "amazingly smooth" it's texture was. 
