$ datamash -t, count 3 sum 3 < file
3,604720
Alguns testes
$ time gawk -F',' '{ sum += $3 } END{ print sum, NR }' longfile
604720000000 3000000
real 0m2.851s
user 0m2.784s
sys 0m0.068s
$ time mawk -F',' '{ sum += $3 } END{ print sum, NR }' longfile
6.0472e+11 3000000
real 0m0.967s
user 0m0.920s
sys 0m0.048s
$ time perl -F, -nle '$sum += $F[2] }{ print "$.,$sum"' longfile
3000000,604720000000
real 0m3.394s
user 0m3.364s
sys 0m0.036s
$ time { cut -d, -f3 <longfile |paste -s -d+ - |bc ; }
604720000000
real 0m1.679s
user 0m1.416s
sys 0m0.248s
$ time datamash -t, count 3 sum 3 < longfile
3000000,604720000000
real 0m0.815s
user 0m0.716s
sys 0m0.036s
Portanto, mawk
e datamash
parecem ser a escolha do grupo.