Thread scaling

[kaipartmann]

The multithreading performance could be slightly improved by optimising the summation of the force densities calculated in the individual threads.

Performance comparison:

nthreads	time old [s]	time new [s]	gain
1	284.61	293.12	0.97
2	153.14	150.98	1.01
8	49.91	46.61	1.07
12	43.71	38.6	1.13
16	41.72	35.93	1.16
24	57.4	37.95	1.51
32	98.26	49.7	1.98