REPL startup: the phillips commission report

Scala version: 2.9.0.r23890

Command template, an accurate proxy for repl startup time:
% JAVA_OPTS="$JAVAOPTS" time scala $SCALAOPTS -nocompdaemon -e '5*5'

$JAVAOPTS is some combination of -d32/-d64 and -client/-server
$SCALAOPTS is --usebootcp or ""

Option --usebootcp places the scala jars on the boot classpath, where
the TSA ushers them through without bothering to verify the absence of
malicious intent. Times were collected in bunches, but they were plenty
stable so the reported times are single representative runs. All
collected on my macbook pro.

[1] OSX Java SE (build 1.6.0_22-b04-307-10M3261)
[2] OpenJDK 32-bit (build 1.7.0-internal-henri_2010_12_01_00_46-b00)
[3] OpenJDK 64-Bit (build 1.7.0-internal-henri_2010_12_01_00_49-b00)

(For [2] and [3] see http://code.google.com/p/openjdk-osx-build/ )

JVM | 32/64 | c/s | bootcp | startup time
-------------------------------------------
[1] -d32 -client --usebootcp | 2.56 real 3.10 user 0.20 sys
[1] -d32 -client | 3.22 real 3.86 user 0.21 sys
[1] -d32 -server --usebootcp | 3.70 real 7.40 user 0.29 sys
[1] -d32 -server | 4.57 real 9.04 user 0.33 sys
[1] -d64 -server --usebootcp | 3.72 real 7.31 user 0.41 sys
[1] -d64 -server | 4.52 real 8.68 user 0.44 sys
[2] -d32 -client --usebootcp | 2.31 real 2.57 user 0.14 sys
[2] -d32 -client | 3.13 real 3.40 user 0.15 sys
[2] -d32 -server --usebootcp | 3.53 real 6.87 user 0.20 sys
[2] -d32 -server | 4.58 real 8.49 user 0.23 sys
[3] -d64 -server --usebootcp | 3.46 real 7.03 user 0.35 sys
[3] -d64 -server | 4.36 real 8.42 user 0.36 sys

TL/DR:
1) --usebootcp drops 0.8-1.0 seconds off startup.
2) -client drops about 1.5 seconds compared to -server.
3) 64bit is not slower than 32bit when comparing -server vs. -server.
4) OpenJDK is faster than 1.6.0_22 at starting scala repls.

Observation: --usebootcp should be the default, and --nobootcp the option.