Store-to-load forwarding for two store/load overlap patterns

Scenario A: a single 8-byte store writes address p, and the next instruction loads the upper 4 bytes at p+4, while the store is still in the store buffer. Scenario B: two separate 4-byte stores write p and p+4, and the next instruction loads all 8 bytes at p, again before either store has drained. On mainstream x86 microarchitectures (Sandy Bridge and later), how do these two loads differ?

main.cpp
// Scenario A: one 8-byte store, then a narrower load contained within it
*p64 = 0x1122334455667788LL;
int32_t a = *(reinterpret_cast<int32_t*>(p64) + 1);   // upper 4 bytes

// Scenario B: two 4-byte stores, then one load spanning both of them
p32[0] = 0x55667788;
p32[1] = 0x11223344;
int64_t b = *reinterpret_cast<int64_t*>(p32);          // all 8 bytes

16 of 47 in Computer Architecture