Correct me if I'm wrong, but as far as I remember:
COLL = Orient and permute corners in one step, without messing up edge orientation
CLL = Same as COLL, but with side-effects on edge orientation
EPLL = Permute edges in one step (and is actually a subset of PLL)
ELL = Permute and orient edges in one step
COLL/EPLL is only usable in a solve where the edges are already oriented at the end of F2L, such as VH, ZZ or ZB.
If edges are not completed, then it would be better to do CLL/ELL, since preservation of edge orientation in COLL would just be wasting moves. In this case you might as well just do OLL/PLL since case recognition is thought to be easier than CLL/ELL.
Now to answer the question: Assuming you are working with a method which orients the LL edges during F2L:
I don't have the exact stats on this, but as far as I remember the average move count of OLL and COLL are about the same (around 9 moves).
COLL reduces the final edges permutation stage to 4 simple PLL cases (Ua, Ub, H and Z, aka EPLL) which has a lower average move count than full PLL, plus there's a better chance of of PLL skip.
Using OLL/PLL means that only the corner orientation cases of OLL need to be used, which again reduces the move count on the OLL phase.
Because I don't know the exact stats I'm not sure whether COLL/EPLL or OLL/PLL has a lower move count. However, it could be argued that case recognition for COLL/EPLL is easier than OLL/PLL, since more of the non U-face stickers are on the U-face for COLL. Recognition of the 21 PLL cases requires more looking round the sides of the cube than EPLL.
So Is COLL/EPLL faster than OLL/PLL? Well, move count is similar and recognition is easier, so in theory yes. But I only know OLL/PLL so don't take my word for it
Does anyone have the average move counts for OLL/PLL and COLL/EPLL, it'd be interesting to compare??
