IME, flac compression for 24 bit signals is only about 0.7 vs about 0.5 for 16 bit data using compression level 6. On the fly will likely use an even lower compression level so don't expect this to double your 24 bit recording capacity.
I've had much better compression ratios for 24 bit audio with FLAC v1.1.4, relative to the older version. v1.1.4 was just released a few months ago, so make the switch if you haven't already. I wonder if SD is going to use the latest version of FLAC with the recorders...
libFLAC still has some problem handling some types of 24-bit data as has been noticed by some of the developers using FLAC. I have seen some posts here by tapers indicating that also. I'm going to include Josh Coalson's recent post on this subject. Josh is the lead developer on the FLAC project.
I think that it's great that SoundDevices is moving in this direction. And it sounds like this issue is going to have a published fix available soon anyway.
Subject: Re: [Flac-dev] FLAC 24 bit test results
From: Josh Coalson <xflac@yahoo.com>
Date: Sun, 8 Apr 2007 15:29:40 -0700 (PDT)
To: flac-dev@xiph.org
>sorry about the delay in responding. Miroslav basically described
>the problem correctly, 24-bit samples with a lot of noise will
>require a rice parameter larger than FLAC supports, so it will punt
>and use VERBATIM subframes. the escape parameter will help a bit
>but is not a good long term solution.
>
>the proper fix is to increase the possible size of the rice parameter
>in a backward-compatible way and I have already done this months ago,
>but I am still working with all the hardware vendors to ensure there
>are no problems for users because it requires a change to the decoder.
>the next flac release will have the change in the decoder but I am
>holding off the encoder release until hardware makers are up to date.
>this should not be too long as very few support 24bps.
>
>so, short answer, the fix is in progress.
>
>Josh