Gigabit ethernet signals through a custom connector?

Hi All,
I'm working on a small, portable, handheld device that is required to
have gigabit ethernet available to a user for certain use-cases.  In
other words, this device will be used most of the time without the
ethernet powered up.  However, when ethernet is needed, I'll need
gigabit.  Since the device is very size-constrained, we'd like to
include as little of the actual ethernet hardware lineup (RJ45,
magnetics, phy) as possible on the actual device, opting for a plug-in
dongle that interfaces with the unit to provide ethernet connectivity
(RJ45 + whatever other hw is required) when needed.  As a side-note, I
was quite surprised to see that there didn't seem to be any slim form-
factor connectors that supported gigE...after all, with SATA etc.
becoming so prominent I figured there would be a comparable small, low-
profile connector for ethernet.  But I can't find one.  So we're
looking at using a low-profile, shielded connector with sufficient
lines to pass the high-speed signals between our device and the
dongle.  We're still looking around, but it looks like Amphenol and
Samtec both have options that may work here.

So this is where things get a bit tricky.  Gigabit ethernet signalling
can be fairly finnicky.  We are exploring the options of where we
could possibly split the hardware between the main device and the
dongle.  The interface between the MAC and the PHY (for RGMII) is ~12
single-ended digital lines running at 125 MHz (but the data lines are
DDR, so these 8 lines are really running at 250 MHz).  The interface
between the PHY and the magnetics is 4 analog differential pairs, with
each pair operating at 125 Msymbols/sec.  The interface between the
magnetics and the RJ45 is also 4 analog differential pairs (they are
isolated, due to the transformer).

Does anyone have any wisdom/experience they would like to share
regarding the least risky spot to drop a connector into the
aforementioned lineup?  One major feather in our cap is that we can
control all parameters of the dongle (i.e., its length, the impedance
matching, the cable type used, etc).  This device is only intended to
be plugged directly into a PC, so it isn't like we'll need to support
the typical 100 meter requirements that IEEE specifies.  Personally,
I'm leaning towards making the split between the PHY and the
magnetics, and passing the four analog differential pairs through our
low-profile connector to the dongle.  But I'd be interested in other
opinions/experiences here.

Thanks for the bandwidth,
John