...there several way to position a buffer on a blender...
Can someone points me out the pros/cons of each design ?
There's surprisingly very little difference in any of those schematics, actually. The main thing we should be concerned with is the load presented to the bass itself - the input impedance.
In the first two cases the topology is virtually the same and the input impedance is identical - 1M. This is literally 'set' by the 1M resistors. The fact that the op amps inputs are in parallel makes very little difference to the input impedance. (Impedance of the non-inverting terminal of an op amp is gigantic and as such is usually assumed to be infinite.)
In the buff'n'blend we still have resistors setting the input impedance - they recommend between 1 and 10M, so depending on what you select you would be able to get very similar results as the two above. Again, the input impedance of the FET is very high and would have minimal impact on the signal from your bass.
In the paralooper, again we have a 1M input resistor feeing into the massive input impedance of an op amp, so it really is very much the same again. The fact that this only has one op amp means virtually nothing because the 1M resistor's affect on the signal at that point is much, much greater than anything the op amp(s) might contribute.
So, from that perspective they are virtually identical.
If you want my opinion, the paralooper is the "right" way of doing this - a good op amp will happily drive both signal paths so there's really no need to have two op amps just to split the signal. The difference with the paralooper is that U2-B and it's associated components forms a low-pass filter, whereas the other designs pass full range signal at this point.
The buff'n'blend is simple, but lacks a buffer for the 'return' signal which may cause difficulty with some effects combinations.