It's area 25, number of effects 5. But the effects all strike together. In each "effect", the aoe will not overlap. But the aoe from multiple effects can overlap.
Like this (the log is in "typescript", one cast of Legions in the log):
# grep "aoe1 dmg64 eff10" typescript
blastsqr: unr2034 x21 y15 aoe1 dmg64 eff10 spc1086324736 as10193 al9
.... etc ...
# grep "aoe1 dmg64 eff10" typescript | wc -l
125
# grep "aoe1 dmg64 eff10" typescript | head -25 | sort -u | wc -l
25
# grep "aoe1 dmg64 eff10" typescript | sort -u | wc -l
48
This is with five clusters of army units; the same single cluster was chosen for a "hit" by all 5 "effects". All five clusters were in range, centered about 6 squares away, but max distance from each other. Only the units in one cluster got affected.
Legions of Steel could be just the equivalent of about an aoe-50 buff, for scoring-calculating purposes.
And Legions does rank below some self-buffs, depending on who's casting. A Marverni druid preferred to spam Legions.
A A3D3S3F3E3N3W3 pretender preferred to self-buff first:
spellscore, Invulnerability score 41579 (boost 99 scorat 0)
spellscore, Astral Shield score 36663 (boost 90 scorat 0)
spellscore, Fire Shield score 18540 (boost 91 scorat 0)
spellscore, Legions of Steel score 15044 (boost 116 scorat 0)
And the best part is, none of this analysis help in the slightest for stopping ordinary E2+ casters from spamming it.
The other "best" part is during testing I finally ran into Satima and her fancy robe, first time ever ... grrr...