It's not about storage "space" in the FW, it's about memory "space" for the video output being sent to the HDMI port. It's generated faster than the HDMI port can receive it and because there's no frame buffer it has to use a few tricks, almost like a real-time frame assembly buffer:
It's about 25minutes in.
While it is a buffer of sorts, it's not a frame buffer because it's already sending the frame data to the HDMI port before the frame is completely generated, like an Ethernet switch that starts routing the Ethernet frame as soon as it get's the MAC address and before it even finished receiving the frame. The video frames are still coming out in real-time as fast as HDMI allows.
It seems that 4.5x is the limit for this trick. The issue here is that Kevtris thinks people want 5x240p=1200p when they actually want 5x216p=1080p (cropped), so he's not even considering their request. A 1080p 5x mode is perfectly do-able and wouldn't take any more resources than 4x 1080p. The cropped data does not need to be manipulated at all, which frees up the resources required. This would allow for integer scaling with an approximation of the original non-square pixels and the cropped area would not be more than typical overscan that was normal when those games were made.