Would the Mercury availability figure be calculated that way, or is that just how you are estimating it because you don't have access to the actual data on how long each geothermal station was offline during the year?
I think geothermal generation is slightly seasonal: higher production in cold winter months than in hot summer months. If you calculated availability just by how long the station is offline in a year then that would not be affected by when in the year it was offline. But if you calculate from the total GWh output for the year as you are doing then you might get different results depending on whether the outage was in winter or summer. I don't know if that would be enough to account for the discrepancy though.
Edit: OK I see in the 2016 FY results presentation Mercury claim 95.5% availability, which is equivalent to downtime of 16.5 days out of 366 maximum. But your calculation is in terms of energy, i.e. a loss of 95.84 GWh out of 2925.84 GWh maximum. So to convert the energy figure to days of downtime would depend on whether the downtime was in winter or summer. A shorter winter outage might lose the same amount of power as a longer summer outage. (Due to the lower thermal efficiency during summer.)
And to complete the process, although there will likely be a large error due to rounding: 95.84 GWh / 16.47 days = 242 MW, so it seems to be in the realm of possibility that your calculation of 96.7% on a lost energy basis does work out equivalent to Mercury's 95.5% availability on a downtime basis, with the downtime occuring mainly during the summer months.