Current performance results for the Pass8 code are on this page .
For 20,000 event clumps:
job step
average CPU time
processClump
183 min
mergeClumps
10 min
Plot of job throughput as of 20130813

From the plot, one can see a processing rate of ~208 runs/day. Assuming 28500 total runs, this would mean a repro time of 137 days or 4.5 months.
Another plot showing the number of run scratch directories cleaned up per day:
Image Added

P300x-FITS
Anchor
P300-FITS
P300-FITS

...

<ac:structured-macro ac:name="unmigrated-wiki-markup" ac:schema-version="1" ac:macro-id="d6dfe3e9e68a8eef-639ec926-46f14d98-b50c9f4c-cc83eb2138b44b849bf28603"><ac:plain-text-body><![CDATA[	Data Product	destination	data content [1]	event selection [1]	makeFT1	gtselect	gtdiffrsp	gtmktime	]]></ac:plain-text-body></ac:structured-macro>
EXTENDEDFT1	SLAC	FT1variables	((FT1EventClass & 0x00003EFF)!=0) pass7.6_Extended_cuts_L1
FT1	FSSC+SLAC	FT1variables	'source' and above EVENT_CLASS bits 2,3,4 evclass=2 filtered from EXTENDEDFT1			(inherited)
EXTENDEDLS1	SLAC	LS1variables	((FT1EventClass & 0x00003EFF)!=0) pass7.6_Extended_cuts_L1
LS1	FSSC+SLAC	LS1variables	'transient' and above EVENT_CLASS bits 0,2,3,4 evclass=0 filtered from EXTENDEDLS1			(inherited)
ELECTRONFT1	SLAC	FT1variables	CTBParticleType==1 pass7.6_Electrons_cuts_L1

...

Space shortcuts

job step	average CPU time
processClump	183 min
mergeClumps	10 min