I only use 50 or 100ug in the 8uL reaction. Only using 1uL or 2uL transformed into 10uL or 20uL of competent cells will yield me between 50-200 colonies (dependent on the destination vector). My negative controls generally do not form any colonies, though if you do a larger transformation (such as 4uL in 40uL), some colonies do show up in the negative control, though you will also have more colonies overall. I know some who scale down their LR reactions to 2uL total and then transform all of it. It seems to work fine for them, but I have not tried it. It should work very well as long as your recombination sites are intact and you still have the ccDB gene in the destination vector (sometimes they can be lost in propagating your vectors and you don't screen properly).

Thanks for your nice experiences, it seems that the efficiency of Gateway is very condition-dependent. Maybe I should test its efficiency on my own condition.