I think vinyl silanes are quite prone to protodesilyation (see e.g. Tet Lett 1995, 9505), although I'm not sure whether in this case the palladium is involved or not. It wouldn't be unreasonable for the NEt3H+ formed in situ to be acidic enough to protonate the C-Si bond, maybe assisted by nucleophilic attack of I- at silicon. I'm not sure how likely your hydropalladation-elimination proposal is - I think once H-Pd-I forms it is pretty rapidly reduced by NEt3 to Pd0, but I could easily be wrong on that.
Does the minor (desired) product convert to the major one over time? Or does the product distribution remain constant?
You could probably solve the problem by using e.g. TBS-ethylene, but sounds like you've got around it by making the conditions a little more gentle anyway.