On the page of Mathworks I found a description of their pivoting strategy for symmetric matrices (http://de.mathworks.com/help/matlab/ref/lu.html):
If A is a square matrix with a mostly symmetric structure and mostly nonzero diagonal, MATLAB® uses a symmetric pivoting strategy. For this strategy, the diagonal where
A(i,j) >= thresh(2) * max(abs(A(j:m,j)))
is selected. If the diagonal entry fails this test, a pivot entry below the diagonal is selected, using thresh(1). In this case, L has entries with absolute value 1/min(thresh) or less.
Unfortunately, I can not figure out what this exactly means, but I would like to use this strategy as it seems to work very well, considering the additional fill ins. The original problem is a quite big, nearly symmetric, complex sparse matrix, which I would like to decompose. With partial pivoting I always run out of memory. Matlab is able to this decomposition with a memory footprint of roughly 50 MB, using presumably the strategy mentioned above.