Actually SMD packages can be surprisingly resilient against G forces. They have short leads and low overall mass so they don't tend to cause bending under acceleration as much. Things like QFN packages have no leads at all even so there is nothing to break off. Tho when going that route it becomes very important how the PCB is mounted as acceleration might cause it to flex and bend and that can really stress the joints under QFNs and cause it to fail.
Mechanical PCB design is actually very important with MEMS sensors. Not only because they almost always come in QFN like packages but putting any bending forces on the package can affect the readings of the sensor. Also if the PCB developes any mechanical resonances, they could cause a large amount of extra acceleration on the sensor, that gets picked up as extra noise or offsets in the readings.
Adding to Berni's comments, MEMS devices can be damaged by seemingly innocuous physical shocks. We experienced low manufacturing yield on a device with an accelerometer. There were several boards to a panel, all connected via mouse bites. When depaneled by hand, sufficiently high forces were generated to damage the MEMS. The datasheet said these chips could survive 10,000G so it seems as though the "snap" of FR4 is capable of generating even more than that. We changed the depaneling procedure and the yield went up dramatically.
I doubt a race car wheel will produce 10,000G, but you'll want to make sure the sensors are working to spec after being manufactured, and you'll probably want to test every one individually.
So, how do you test a 50G sensor when it'll only read 2% of full scale sitting on the bench?