SLSQP with L-BFGS update

[AccessViolationEnjoyer]

Is there any particular reason why the SLSQP method uses the full BFGS update rather than the limited memory update? It seems to perform well for me on difficult problems with a small number of variables, but quickly becomes less efficient that the L-BFGS-B method, for example, when tackling problems with 50 or so variables. I'm guessing that the reason for this is the O(n^3) hessian approximation update.