Hi Thomas,
Between RBF and linear scaling, you would really need to compare muscle moment arms. I think the origin, insertion, and via points for the muscles will scale differently between the two scaling methods and a small change in the moment arm can lead to large change in the muscle force.
Regarding the higher second peak, this can be due to a few reasons. At toe-off, the gastrocnemius probably shares more of the load than it should, resulting in higher knee reaction forces. This can be due to more advantageous moment arm of the gastrocnemius compared to soleus. You can probably fix this by passing them both through a common via-point. Another reason could be the way the muscles are split. Soleus is split into multiple units and this affects how muscles are recruited. See this link. You can either replace the soleus units with a single soleus muscle or you can introduce a normalization factor in each of the muscles. See this previous thread on how to do it.
At this point, I would recommend you consider upgrading and requesting access to the beta version. The beta version of AMS has new features that can make it easier to normalize based on muscle volume.
Best regards,
Dave