Alright, some data 
Rouse = +40%
Fey Illumination = +20%
If additive, we expect 60% as combined value, if multiplicative, we expect 68%.
20 non-crit samples without buffs and with both buffs:
Code:
base rouse + fey illumation
2030 3424
2007 3424
2024 3465
2016 3320
2049 3474
1949 3285
2008 3389
2015 3460
2031 3444
1968 3470
2010 3462
1971 3376
1952 3276
2048 3368
2032 3407
1959 3372
2038 3351
2049 3432
2002 3389
2049 3301
Average:
- Base: 2010.35
- Rouse + Fey Illumation: 3394.45
3394.45/2010.35 = 1.6885
So +68.85%, a strong indication it's multiplicative.
Still not entirely a proof.. and a statistical test requires more data as the data isn't normally distributed, but I think this is enough to assume it's multiplicative.
Also @ nekotee.. yes, it appears probability-related stats do stack additively. Which makes sense from a design perspective.. It makes all those buffs situation-independent, making it easy to reason about them.