Interesting - I almost needed to pinch myself. Luckily I had a Moteino with regulator removed and fuses set to no BOD lying around. As I said with this setup at 17dBm the receiver saw the test packages sent 2.2V at -34 RSSI in the beginning. From there is just continued down, very reliably, less than 0.3% packet loss.
It continued all the way down to 1.69V when the 328p just couldn't keep up appearances and suffered from ram corruption. Until then the radio hadn't missed a beat:
Apr 22 22:03:18 espgw.lx [11] [RX_RSSI:-36,1,8,45]
Apr 22 22:03:18 espgw.lx test4 -> nd:11,nr:285,pw:12,vc:170,rt:3
Apr 22 22:03:21 espgw.lx [11] [RX_RSSI:-37,1,8,45]
Apr 22 22:03:21 espgw.lx test4 -> nd:11,nr:286,pw:12,vc:170,rt:3
Apr 22 22:03:24 espgw.lx [11] [RX_RSSI:-35,1,8,45]
Apr 22 22:03:24 espgw.lx test4 -> nd:11,nr:287,pw:12,vc:170,rt:3
Apr 22 22:03:27 espgw.lx [11] [RX_RSSI:-36,1,8,45]
Apr 22 22:03:27 espgw.lx test4 -> nd:11,nr:288,pw:12,vc:170,rt:3
Apr 22 22:03:30 espgw.lx [11] [RX_RSSI:-35,1,8,45]
Apr 22 22:03:30 espgw.lx test4 -> nd:11,nr:289,pw:12,vc:170,rt:3
Apr 22 22:03:33 espgw.lx [11] [RX_RSSI:-38,1,8,45]
Apr 22 22:03:33 espgw.lx test4 -> nd:11,nr:290,pw:12,vc:170,rt:3
Apr 22 22:03:36 espgw.lx [11] [RX_RSSI:-35,1,8,45]
Apr 22 22:03:36 espgw.lx test4 -> nd:11,nr:291,pw:12,vc:170,rt:3
Apr 22 22:03:39 espgw.lx [11] [RX_RSSI:-35,1,8,45]
Apr 22 22:03:39 espgw.lx test4 -> nd:11,nr:292,pw:12,vc:170,rt:3
Apr 22 22:03:45 espgw.lx [11] [RX_RSSI:-38,1,8,45]
Apr 22 22:03:45 espgw.lx test4 -> nd:11,nr:294,pw:12,vc:169,rt:17
Apr 22 22:03:51 espgw.lx [11] [RX_RSSI:-38,6,8,45]
Apr 22 22:03:51 espgw.lx test4 -> nd:11,nr:296,pw:12,vc:169,rt:30
Apr 22 22:03:54 espgw.lx [11] [RX_RSSI:-37,1,8,45]
Apr 22 22:03:54 espgw.lx test4 -> nd:11,nr:297,pw:12,vc:169,rt??.
Apr 22 22:04:00 espgw.lx [11] [RX_RSSI:-38,1,8,45]
Apr 22 22:04:00 espgw.lx test4 -> nd:11,nr:299,pw:12,vc:169,rt:33
Apr 22 22:04:06 espgw.lx FreeHeap: 17440
Apr 22 22:04:06 espgw.lx [11] [RX_RSSI:-38,1,8,45]
Apr 22 22:04:06 espgw.lx test4 -> nd:11,nr:301?test4 -> ??#014?B?00?WZ6Q^??
Apr 22 22:04:12 espgw.lx [11] [RX_RSSI:-36,1,8,45]
Apr 22 22:04:12 espgw.lx test4 -> nd:11,nr:303,pw:12,vc:169,rtNYV
Apr 22 22:04:15 espgw.lx [11] [RX_RSSI:-36,1,8,45]
Apr 22 22:04:15 espgw.lx test4 -> nd:11,nr:304,pw:12,vc:169,rtNYV
Apr 22 22:04:43 espgw.lx [11] [RX_RSSI:-38,6,8,45]
Apr 22 22:04:43 espgw.lx test4 -> nd:11,nr:313I?ج?"?A?k???t]#021
Apr 22 22:04:46 espgw.lx [11] [RX_RSSI:-35,6,8,45]
Apr 22 22:04:46 espgw.lx test4 -> nd:11,ns:314,pw:12,vc:168,rt???F
Apr 22 22:04:55 espgw.lx [11] [RX_RSSI:-35,6,8,45]
Apr 22 22:04:55 espgw.lx test4 -> nd:11,ns:317?-r:2??#006?#6???.
Apr 22 22:04:55 espgw.lx [11] [RX_RSSI:-36,1,8,45]
Apr 22 22:04:55 espgw.lx test4 -> nd:11,ns:317?o*'`???#013\??!?|a#021??K
RSSI pretty much unchanged at -36 vs -34 at 2.2V.
So I guess this should put the hypothesis that the rfm69hw only works down to 2.5V to a conclusive rest.