Using sample path analysis we show that under the same load the mean d
elay in queue in the M/G/2 system is smaller than that in the correspo
nding M/G/1 system, when the time has either the DMRL or NBU property
and the service discipline is FCFS. The roof technique uses a new devi
ce that equalizes the work in a two server system with that in a;singl
e sterver system. Other interesting quantities such as the average dif
ference in work between the two servers in the GI/G/2 system and an ex
act alternate derivation of the mean delay in the M/M/2 system from sa
mple path analysis are presented. For the same load, we also show that
the mean delay in the M/G/C system with general service time distribu
tion is smaller than that in the M/G/1 system when the traffic intensi
ty is less than 1/c.