Always a very interesting debate.

Frankly very little evidence exists for mobilisation of true Wikipedia reference-linkfrozen shoulders - and what does exist lacks robustness and any valuable methodological detail. If you mobilise the cervical/throracic spine as suggested you may slightly functional ROM/pain levels but actual passive glenohumeral joint will not change if it is true frozen shoulder. At times this may be a viable option but in my opinion this should only take 1-2 sessions to achieve the desired gains.

Overall I have to agree with sande's approach, particularly from stage 2 onwards in that home/exercise/activity modification should form the mainstay of treatment - doing otherwise is over-servicing.