Probably Heise when solving for speed and starting on any color.

It's not too hard to use <45 turns with it. 38 could probably be done too with lots of practice. HTM. If you count STM, then non-matching Roux (using the same L/R colors every time) can be equal to Heise. I use 43 on average using it.

I think the Petrus method is pretty good at getting less turns...I may be wrong though.

It seriously depends on what you're talking about.

The method that uses the least number of moves is the optimal sequence.

Take a look at Dan Harris's Fewest Moves Challenge. A lot of people use a variant of Petrus method. I, when I still had time to do something like that, used what's known as a 2-gen method: 2x3x3, orienting edges and permuting corners, and then finally the rest of the cube with solely 2-gen. It certainly does not work efficiently for a speedsolving method.