Most conducted and radiated EMI is caused by common mode noise on signals entering or exiting a box. If you tie your box ground to digital ground with a very low DC and RF impedance (large, fat and very short connections) at the point a signal exits or enters the box the common mode noise is significantly reduced. On the power leads a capacitor (e.g. 470nF) with a low RF ground connection to chassis (short, fat connections) tends to get rid of a lot of power-line conducted EMI. A 5mm wire will have significant impedance at RF frequencies and will do very little to stop it, you need wide, fat and very short connections.
Mark.
1. Just how fat and how short does it need to be?
If 5mm is already too long, then I guess that by itself might explain why the 47-ohm antenna "short" I tried seemed to work so poorly.
2. Unrelated: Since the RFM69's radio is being driven by a 32Mhz crystal, does that make it more likely that the RFM69 could be getting EMI/RFI directly from the rest of the Moteino, which, IIRC, typically operates at 8Mhz? I guess the hypothesis would be that whatever up-converts the 32Mhz crystal frequency up to 915Mhz might be upconverting whatever noise/harmonics there might be also.
Maybe a ghetto way to test the hypothesis would be to see whether the Moteino can receive a distant node while the RFM69 is in Listen-Mode, with the rest of the Monteino asleep, that it can't otherwise receive with the Moteino fully powered on? Such a test would be easier said than done, I realize. Perhaps someone reading this has proper instrumentation and could check for it directly without resorting to ghetto methods.
However, if it turns out to be true, then it would be good to know: for instance, maybe running a modified Moteino at a much lower speed, like 1Mhz, would yield reduced EMI/RFI? If such a modification were made, it could at least be tested for effect using the RSSI measurements being discussed on the other thread.