Large-but-correctly-aligned-and-optimized code is faster than less-bytes-per-instruction/opcode-packed code

Is large-but-correctly-aligned-and-optimized code faster than less-bytes-per-instruction/opcode-packed code?

Alex Ionescu mentioned in ros-dev mailing list:

I’m not sure why you would want kernel code to be “smaller” instead of “faster” though — on modern processors for cases like interrupts and such, large-but-correctly-aligned-and-optimized code is faster than less-bytes-per-instruction/opcode-packed code.
ie:

mov eax, [foo]
add eax, 1
mov [foo], eax

is faster than

inc [foo]