C-----*-----3.3V
Gen--50R---B      |
             E     10nF
             |     |
             1k    gnd
             |
            gnd

measuring Vb and Ve. Transistors: 2N2222a and BFR93a.

BFR93a:
Generator is adjusted to Vdc=2.3V and 5Vpp.

At 1 MHz, the emitter signal waveform is a nice trapezoidal signal with
a steeper slope when leaving the upper clipping level. The simulator
predicts rounded tops and a somewhat upwards sloping bottom.
At 27 MHz the Spice model predicts negative signal excursions. I could
not observe them. But if I add some 5-10 pF at the emitter resistance to
account for the probe, simulations and reality almost match.

2N2222a:
Generator is adjusted to Vdc=0 and 5Vpp.

At 27 MHz, Spice model predicts sinusoidal signal centered around 0
approx +-2.3. On the scope I see max +1.8 and min -1.5, i.e. there is
some loss but the qualitative shape is ok. Of course, this is not what
the circuit is expected to do, but indicates the model is ok.
At 1 MHz, Spice model predicts bottom clipping but with positive slope
of 1.9V/us. Measurements more or less confirm this.

At 27 MHz, a quite sinusoidal output signal of 4Vpp can be measured
adjusting Vdc=1.3. The simulator predicts the same, although with a
4.5Vpp amplitude. So, the model seems ok.

Conclusion:

The Spice models are ok and they do model large signal behavior. It was
my error to assume that there could be no negative signal excursions:
this is only true at low frequencies. Somehow I thought that 27 MHz was
dc for a 5 GHz transistor. But his is not the case for a high input
impedance circuit, as JT pointed out.

Thanks to all.

Pere