in some cases pyramid friction approximation is too
inaccurate (eg. bodies are harder to push along
diagonals (between fdir1 and fdir2). with large mu, mu2
values and high body inertia, body might even not move
along diagonals. adjusting fdir1 is inefficient because
it's impossible to predict in which direction external
forces will move body).
full friction cone might be slow but I would use it
only for some, not all bodies. CPUs are getting faster
and faster so full friction cone could finaly appear as
not being so slow when used for several contacts.