I'm with @weasel79 on skepticism with having the full Linux stack. There are a couple of problems I see with it. Latency and performance is probably number one. They claim to have solved that with some Linux kernel hacks, etc:
"Bela runs a 4.4 Linux kernel with the Xenomai real-time extensions to provide ultra-low-latency performance. Xenomai Cobalt is a co-kernel for Linux that allows [running] selected threads at hard real-time priority, bypassing the Linux kernel to achieve performance comparable to running bare metal without an operating system. We are using an onboard 200MHz microcontroller (Programmable Realtime Unit) available on the Texas Instrument AM3358 system-on-chip (SoC) to act as a sophisticated DMA [direct memory access] controller, performing the low-level input/output operations for the audio channels (over I2S), analog channels (over SPI), and digital channels (the SoC's GPIOs)."
I haven't worked with it at all so I don't know if their claim of sub 1ms latency is legit. I know from first hand experience as a daily Linux user how painful its audio stack can be. I'm guessing that this depends on exactly what you're doing. The apps you're running on Bela might not actually respect that deadline because they're used to running on a normal Linux desktop with the entire audio stack (aside: it sounds like some of them are patched or have various Bela-compatibility layers available). This leads to the second point:
Because you just have a regular Linux box with a bunch of compatible apps, it's not as focused as it could be. It's hard to know what interactions might arise between the various apps and the Linux audio stack and kernel hacks unique to Bela. You're introducing many layers of complexity that might get in the way of just computing samples efficiently and driving a DAC.
On the other hand, we may get to a point where running full Linux might just be fast and stable enough for realtime audio that no one will ever notice. Having a full Linux box is really nice for more advanced stuff like networking, bluetooth, usb, etc. But again is that stuff really needed for something that is just trying to do audio dsp as efficiently as possible on a small piece of hardware?