I would try something like this to test several options.
library(maptools) library(ggplot2) library(rgdal) library(raster) library(latticeExtra)
Download and read data
download.file('https://dl.dropboxusercontent.com/u/46043231/UK.zip', "uk.zip", method="internal", mode="wb") unzip('uk.zip') uk <- readOGR(dsn = getwd(), layer = 'uk_outline_1000m')
It is assumed that OSGB 1936 / British national grid is used.
More on SpatialReference
proj4string(uk) <- CRS('+init=epsg:27700')
Record projected shapefile uk bbox. I will write this to map layers to QGIS. This will be my link system.
writeOGR(as(bb.uk, 'SpatialPolygonsDataFrame'), dsn = getwd(), layer = 'bbuk2_bng', driver = 'ESRI Shapefile')
The required bounding box. Using predicted coordinates
bb.uk2 <- as(extent(c(0, 630000), c(0, 1000000)), 'SpatialPolygons') proj4string(bb.uk2) <- CRS('+init=epsg:27700')
Record projected bbox shapefile
writeOGR(as(bb.uk2, 'SpatialPolygonsDataFrame'), dsn = getwd(), layer = 'bbuk2user_bng', driver = 'ESRI Shapefile')
QGis map using British National Grid EPSG: 27700

Planned Layers
Base plot
plot(uk, col = 'grey50', axes = T, xlim=c(-50000, 705566.4), ylim=c(-50000, 1325000)) plot(bb.uk, add = T) plot(bb.uk2, border = 'red', add = T)

spplot
I took an arbitrary window to expand the area of ββthe graph.
sp::spplot(uk, zcol = 'NAME_ISO', scales = list(draw = TRUE), xlim=c(-50000, 705566.4), ylim=c(-50000, 1325000), col.regions="grey90") + latticeExtra::layer(sp.polygons(bb.uk, fill = NA, col = 'blue')) + latticeExtra::layer(sp.polygons(bb.uk2, fill = NA, col = 'red'))

ggmap with projected layers
uk.df = fortify(uk) # admin bbuk.df <- fortify(bb.uk) # country bbox extent bbuk2.df <- fortify(bb.uk2) # user bbox extent
write it down
p <- ggplot() + geom_polygon(data = uk.df, aes(x=long, y=lat, group=group)) + geom_polygon(data = bbuk.df, aes(x=long, y=lat, group=group), colour = 'blue', fill = NA) + geom_polygon(data = bbuk2.df, aes(x=long, y=lat, group=group), colour = 'red', fill = NA) + coord_equal() # cartesian p

build it using the user bounding box
p + coord_equal(xlim=c(0, 630000), ylim=c(0, 1000000))

Now ggplot with geographic (non-projected) coordinates
Undesigned WGS84 coordinate system
p.wgs84 <- CRS("+init=epsg:4326")
Convert Projection Layer to WGS84
uk.wgs89 <- spTransform(uk, p.wgs84)
Geographic bbox uk
bbuk.wgs84 <- as(as(extent(uk.wgs89), 'SpatialPolygons'), 'SpatialPolygonsDataFrame')
Geographic bbox user degree
bbuk2.wgs84 <- spTransform(bb.uk2, p.wgs84) bbuk2.wgs84 <- as(bbuk2.wgs84, 'SpatialPolygonsDataFrame')
Build it with ggplot and cartesian map
uk.df = fortify(uk.wgs89) # admin bbuk.df <- fortify(bbuk.wgs84) # country bbox extent bbuk2.df <- fortify(bbuk2.wgs84) # user bbox extent

The result is not what I would expect. I do not understand why the red square is distorted.