SOme much faster comparators..
Just from one company (linear technologies)
www.linear.com
LT1715, LT1719, LT1712, LT1721, LT1720, LT1711, LT685
LT1395, LT1714, LT1713, LT1016, LT1116, LT1671, LT1011, LT319A
The LT1715 has a prop delay of 4 nano-seconds.
Other folks who sell fast comparators, national semi, texas instruments, analog devices.. you can find lots of em out there but you will pay for the higher speeds. And some of them require smaller supply rails and run at higher power levels.. so there are tradeoffs.. but they exist if you really need them.
It is not clear to me that what you are seeing is not simply the expected prop delay in your circuit for that comparator and I think you are having a hard time explaining it (at least to me anyhow). Some things to keep in mind.
Your function generator has noise on it that you are likely not setup to properly measure.
Your function generator has offsets on its output and one channels will be different than another. If you think you have nulled these out show us how. If you are trying to measure single mV with a cruddy hand held multimeter, you are probably not getting the right answer. All of these things can slightly affect the comparators trip point. And as RON mentioned, the specs only show prop times for specific overdrive voltages so if you are not providing these amounts (including none), you can expect to find a prop time that will show up nowhere in the spec.
Only _generally_ speaking can we tell you that the more overdrive you give, the faster it will switch. If you do less, it is anyones guess as to what you might see.
The other fun thing about all this is if you put a different part in there but same type & part #, the times you observe will be different still. Put a heat gun on it and watch the times change again.. maybe not by alot but sounds like you are trying to pin point a single cause with some amount of precision.
Why are you doing all this anyways, if I may ask?