Is there a db-backup related speed bottleneck?

[sb10]

rm ~/.wr_production/db && s3cmd -c ~/.s3cfg.sb10 rm s3://sb10/wr_backups/db_bk
wr manager start -s lsf
cat longmult_inputs | perl -ne '($am, $bm) = split("\t", $_); print "perl ./longmult_step1.pl $am $bm\n"' | wr add -i longmult_step1 --cwd_matters -o 2

Got up to ~150 (without fluctuating down) running simultaneously.
But very slow? Much faster with backups disabled? Try the new coreos version of boltdb? Try the db on ram disk?

It currently (with or without backups) takes 22s, which looks bottle necked vs
33s given 24 -> 80..150
Updating to bbolt made no difference to performance.

wr manager stop && rm ~/.wr_production/db && s3cmd -c ~/.s3cfg.sb10 rm s3://sb10/wr_backups/db_bk
wr manager start
cat longmult_inputs | perl -ne '($am, $bm) = split("\t", $_); print "perl ./longmult_step1.pl $am $bm\n"' | wr add -i longmult_step1 --cwd_matters -o 2 -r 0

Seemed to run fine, but then it abruptly hung with 24 runners going but not completing. wr status "could not reach the server". It never recovered. manager couldn't be stopped. ~/.wr_production/.db_bk_mount/sb10/ could not be lsed without also hanging. Did fuse just wedge?