Just thought I'd add my experience here, given that I have a Billion Bipac 8800NL and saw your post both on this forum and Billion's.
I have noticed very similar behaviour on my router - like you I have a fairly marginal ADSL line so Billion routers are a very good option with their target SNR margin tweaks, which are invaluable to stop my exchange giving me stupidly high SNR margins and throttling the sync as a result. I've managed to take our download speed from around 1.8 meg actual to around 2.8 meg actual, with no noticeable instability or dropouts.
However the connection isn't perfect - I've always noticed latency going out the window when large uploads are occurring (Dropbox, iCloud Backups/Photos, etc.), which is a huge problem when trying to use Skype at the same time. At the moment I'm using a global upload limit QoS rule to limit each network client to 70% of the upload bandwidth, and a QoS prioritising rule matching traffic that vaguely looks like Skype on certain PCs.
Upon running the ICSI test I can also confirm similar results - 2800ms uplink buffering.
Whilst I have QoS turned on normally, as mentioned, I have since run the test both with my QoS rules active and with all QoS rules disabled, and the results are virtually identical - around 3 seconds of uplink buffering, hence bufferbloat.
I'm really hoping for some sort of solution from Billion on this, it sounds like fq_codel is an excellent alternative algorithm and there has clearly been lots of work done by kind people independent from router manufacturers to investigate and solve bufferbloat. Why OEMs can't integrate this into their respective firmwares is beyond me. If a massive security issue suddenly became public that affected router firmwares, surely they would try to rush out a fix to all the affected devices? I appreciate bufferbloat is not quite in the same level of urgency but it seems like a relatively simple fix to implement to give certain users a vastly improved experience.