Gas properties of galaxies have changed between code versions

I have been running simulations of isolated galaxies with different normalization of the entropy floor last year. Re-running them with a recent SWIFT version leads to quite different results which seem to be related to different density - temperature distribution of the gas.

"Old" SWIFT version from last year:

Version : 0.8.5
Revision: v0.8.5-1141-g2c577425, Branch: master, Date: 2020-09-11 16:01:08 +0100 

"New" SWIFT version:

Version : 0.9.0
Revision: v0.9.0-685-g096059af, Branch: master, Date: 2021-10-21 15:43:15 +0000

Both simulation sets are compiled with:
Config. options: '--with-subgrid=EAGLE-XL --with-hydro=sphenix --with-kernel=quartic-spline --with-ext-potential=hernquist --with-tbbmalloc --disable-hand-vec --enable-ipo'

The yaml files are identical beside some parameters that were added / removed between versions:

EAGLEStarFormation:
  EOS_density_norm_H_p_cm3:          0.1       # Physical density used for the normalisation of the EOS assumed for the star-forming gas in Hydrogen atoms per cm^3.
  EOS_temperature_norm_K:            8000      # Temperature om the polytropic EOS assumed for star-forming gas at the density normalisation in Kelvin.
  EOS_gamma_effective:               1.3333333 # Slope the of the polytropic EOS assumed for the star-forming gas.
  SF_threshold: Zdep
  SF_model: PressureLaw

Stars:
  luminosity_filename: ../../data/photometry

EAGLEFeedback:
  SNII_feedback_model: MinimumDistance
  SNII_energy_fraction_function: EAGLE
  SNII_energy_fraction_use_birth_density: 0
  SNII_energy_fraction_use_birth_metallicity: 0

The EAGLEStarFormation EOS parameters are removed in the new version but I had that set to the same values as the entropy floor anyway, so it should not make a difference and "Zdep" and "PressureLaw" should be the default in the old version. In both cases I run without AGB or SNII enrichment and use a fixed SNII energy of 10^51 erg (SNII_energy_fraction_min: 1.0 and SNII_energy_fraction_max: 1.0).

Here are the phase-space diagrams of the 2 (old and new SWIFT) x 5 (different entropy floor) runs:

Old SWIFT:

New SWIFT:

Some differences are already visible here, but the difference is most obvious in the stellar birth densities (left: old SWIFT, right: new SWIFT):

And finally, this is what the galaxies look like: Visually it looks like as if the new SWIFT version produces more small, dense gas clumps, which would align with the higher densities in the phase-space plots and the stellar birth density distribution.

I have run the same ICs also with different resolutions (1.8e6, 2.3e5, 2.8e4 Msol), with different entropy floor slopes (4/3 and 2) as well as for different star formation laws (pressure law and Schmidt law). These differences to the old code are apparent in all these runs.

Summarising, there is higher-density gas in the new version of SWIFT and (best seen in the run with the highest EOS normalization) some gaps at intermediate densities which can be seen in the phase-space diagrams (for a specific snapshot) but also in the horizontal parts of cumulative birth density distribution (i.e. it is not just affecting one snapshot).

Not sure at the moment what has caused this quite significant change between two code versions.