Try using less time in the plasma. We usually don't go more than 45 seconds in our Harrick plasma chamber. Too much time will damage the surface of the PDMS and weaken the bond.

I used the same method and it works fine for me. Maybe the inlet holes are not well made and so the surface can have some roughness.

If your aim is to couple acoustics into the fluidic channel then you can bond your PDMS to a thin glass slide (which is more reliable than bonding to LiNbO3), and place this on the LiNbO3 using ultrasonic coupling gel, as described in the attached paper.This also avoids wasting LiNbO3 samples when the PDMS channel leaks.

http://scitation.aip.org/content/aip/journal/jap/111/9/10.1063/1.4711101

Well, though I saw this topic a bit late nevertheless it is never too late to share knowledge :)

For me, I coated the Lithium Niobate with 100 nm of SiO2 later and then bonding becomes similarly strong as with glass.

Also, keep in mind that this will also give you another advantage that IDTs will be prevented from damage and the shelf-life increases.

Anyways good luck.