The actual problem is a bit more complicated as the shown one. :) I already use as much vectorization as possible. (A(ii,1:end)) I also have a lot of function evaluations which take a long time in mp as well.

Hello.

Thank you for the quick answer. The increasing performance for array operations will give a great overall performance improvement. In my application, 98% of the time is needed to build the system of linear equations Ax=b (A[600,600]) and only 2% for solving it. :-) If I use the double precision from matlab it is like 40% to 60%.

A short further comment:
From Mathematica
N[BesselY[0, 40*I], {1000, 1000}] =
-5.34306132305811469773184318452720415....e-19+1.4894774793419899924224591570721184444946....e16 i
From Matlab
mp.Digits(16)
bessely(0,mp('40*i')) = -2.546479089470325 + 1.48947747934199e+16i
mp.Digits(25)
bessely(0,mp('40*i')) = 7.114780385429948677221705e-09 + 14894774793419899.92422459i

Normal Double Precision:
bessely(0,40*i) = -5.343061323058116e-19 + 1.489477479341990e+16i