Ok, I've done a few tests with imcore with and without compression.  I ran
the same data through imcore, but in 4 different forms.  (1) scaled
integers (as you usually produce) (2) tile-compressed scaled integers,
(3) real (data was re-reduced and not squashed) (4) compressed real with
several differnt values of noisebits.  Here's the info:

datatype   noisebits   filesize   compression   run time  #rows in cat
----------------------------------------------------------------------
integer       --       16789680     1.0           40s       4981
real          --       33579360     1.0           40s       4812
comp int      --        6796800     2.5           40s       4981
comp real      4        6952320     4.8           38s       4811
comp real      6        9048960     3.7           38s       4813
comp real      8       11145600     3.0           39s       4808

int

SKYLEVEL=               403.49 / Median sky brightness (counts/pixel)
SKYNOISE=                13.92 / Pixel noise at sky level (counts)
THRESHOL=                20.89 / Isophotal analysis threshold (counts)
MINPIX  =                    4 / Minimum size for images (pixels)
CROWDED =                    1 / Crowded field analysis flag (0 none, 1 active)
RCORE   =                  3.5 / Core radius for default profile fit (pixels)
SEEING  =                 6.19 / Average FWHM (pixels)

real

SKYLEVEL=               403.60 / Median sky brightness (counts/pixel)
SKYNOISE=                14.10 / Pixel noise at sky level (counts)
THRESHOL=                21.15 / Isophotal analysis threshold (counts)
MINPIX  =                    4 / Minimum size for images (pixels)
CROWDED =                    1 / Crowded field analysis flag (0 none, 1 active)
RCORE   =                  3.5 / Core radius for default profile fit (pixels)
SEEING  =                 6.19 / Average FWHM (pixels)


Basically main difference is caused by nint(sky) and has nothing to do
with the compression algorithm.


int   -v- real

 No. matched within 2.5 arcsec search radius  =  4719

 Average astrometric errors (x,y arcsec)    1.43176E-02    1.36401E-02

 Scatter in mags   =    7.22656E-04

comp4 -v- real

 No. matched within 2.5 arcsec search radius  =  4782


 Average astrometric errors (x,y arcsec)    9.39453E-03    7.22656E-03

 Scatter in mags   =    2.85110E-04

comp6 -v- real

 No. matched within 2.5 arcsec search radius  =  4805

 Average astrometric errors (x,y arcsec)    8.58154E-03    6.59424E-03

 Scatter in mags   =    1.41144E-04

comp8 -v- real

 No. matched within 2.5 arcsec search radius  =  4806

 Average astrometric errors (x,y arcsec)    8.40088E-03    6.32324E-03

 Scatter in mags   =    3.81088E-05