As a prenote: My answer is somehow speculative because I have no experimental data or reference values of similar reactions used. In other words: The argumentation pathway below might be wrong.
Addition of LDA mostly gives the trans-enolate if a primary alkyl substituent next to the carbonyl group has been used.
Carbonyl(O) is smaller than the methyl group [ref.], therefore the allylic A13 interaction of Me//Me or even Me//i-Pr is bigger than the allylic A12 interaction.
An allylic A13 Me//Me interaction is about 3,7 kcal/mol, an allylic A12 Me//Me is may be around 3,2 kcal/mol (this is just a guess).
If HMPA is added Li might coordinate to four HMPA molecules forming a cluster and inhibiting the cyclic transition state of LDA and hence no longer providing the trans-enolate, meaning an open transition state must be lower in activation energy for the cis-enolate formation.
So I tried to draw the conformations of 3-pentanone and then treated the base as "point bulk", meaning the base is like a sphere with different sizes.
Hard bases have a bigger repulsion to the carbonyl oxygen.
I have used arbitrarily numbers for the stereoselectivity.
Here is my answer which base gives more cis-enolate than the other: